Paracrine Activin-A signaling promotes melanoma growth and metastasis through immune evasion

The secreted growth factor Activin-A of the TGFβ family and its receptors can promote or inhibit several cancer hallmarks including tumor cell proliferation and differentiation, vascularization, lymphangiogenesis and inflammation. However, a role in immune evasion and its relationship with tumor-induced muscle wasting and tumor vascularization, and the relative contributions of autocrine versus paracrine Activin signaling remain to be evaluated. To address this, we compared the effects of truncated soluble Activin receptor II B as a ligand trap, or constitutively active mutant type IB receptor versus secreted Activin-A or the related ligand Nodal in mouse and human melanoma cell lines and tumor grafts. We found that while cell-autonomous receptor activation arrested tumor cell proliferation, Activin-A secretion stimulated melanoma cell dedifferentiation and tumor vascularization by functional blood vessels, and it increased primary and metastatic tumor burden and muscle wasting. Importantly, in mice with impaired adaptive immunity, the tumor-promoting effect of Activin-A was lost despite sustained vascularization and cachexia, suggesting that Activin-A promotes melanoma progression by inhibiting anti-tumor immunity. Paracrine Activin-A signaling emerges as a potential target for personalized therapies, both to reduce cachexia and to enhance the efficacy of immunotherapies.


INTRODUCTION
In melanoma research, targeted inhibitors and immune checkpoint therapies are available but their efficacies remain limited by acquired and intrinsic drug resistance mechanisms (Lau et al., 2016). Patients with resistant melanoma have a poor prognosis due to high probability of metastasis, correlating with progression from superficial spreading to invasive vertical growth. Tumor-initiating potential and metastasis also correlate with low immunogenic profiles of subpopulations of cells and phenotypic plasticity marked by pseudo-epithelial-to-mesenchymal transitions (EMT) and the ability to reversibly switch between proliferative and invasive gene signatures (Caramel et al., 2013, Hoek et al., 2008, Schatton and Frank, 2009, Widmer et al., 2012. Thus, elucidating mechanisms of tumor immune evasion and their coupling to cancer cell plasticity is critical to develop effective immunotherapies.
Known local cues in the tumor microenvironment that regulate melanoma cell plasticity and anti-tumor immunity include transforming growth factor beta (TGFβ). In many tumors including melanoma, TGFβ facilitates or inhibits tumor progression, depending on the context (Bellomo et al., 2016, Perrot et al., 2013. In normal melanocytes, TGFβ induces cell cycle arrest and apoptosis (Alanko andSaksela, 2000, Rodeck et al., 1991), but this response is attenuated in cells from benign nevi despite sustained activation of downstream SMAD2,3 transcription factors (Rodeck et al., 1999). SMAD-binding sites in the PAX3 gene mediating repression of pigment synthesis likely contribute to melanoma cell plasticity and phenotype switching (Pinner et al., 2009, Yang et al., 2008. TGFβ immunostaining was found to correlate with the invasive vertical growth phase and metastasis (Van Belle et al., 1996), and blockade of downstream SMAD2,3 transcription factors by overexpression of antagonistic SMAD7 in 1205Lu melanoma cells inhibited tumorigenicity and bone metastasis (Javelaud et al., 2005, Javelaud et al., 2007. In metastatic B16F10 mouse melanoma, TGFβ also reduces Natural Killer cellmediated tumor rejection while promoting the differentiation of immunosuppressive Foxp3+ regulatory T cells in tumor beds and draining lymph nodes (Chen et al., 2003, Gorelik and Flavell, 2001, Turk et al., 2004. However, the functionally relevant active TGFβ in this model is provided by immature myeloid dendritic cells in draining lymph nodes, likely because B16F10 and other melanoma cells themselves secrete TGFβ only in latent form (Ghiringhelli et al., 2005, Yin et al., 2012.
Melanoma cell lines and tumors also frequently express Activin-A (Heinz et al., 2015, Hoek et al., 2006. Activin-A stimulates the same SMAD transcription factors as TGFβ, even though it derives from distinct precursor dimers encoded by the Inhibin βA (INHβA) gene and binds distinct complexes of the Activin/Nodal type I and II receptors Acvr1b/ALK4 or Acvr1c/ALK7 and Acvr2 (reviewed in Hedger et al., 2011, Sozzani and Musso, 2011, Walton et al., 2011. Activin-A can mediate protective or tumorigenic effects (reviewed in Loomans and Andl, 2014). For example in mouse models of pancreatic cancer, Activin-A promotes tumor progression by inhibiting cell differentiation (Lonardo et al., 2011, Togashi et al., 2015. Paradoxically, its receptor ALK4 (ACVR1B) primarily mediates tumor suppressive functions and is frequently deleted in clinical samples (Qiu et al., 2016, Togashi et al., 2014. Cytostatic and proapoptotic signaling by Activin-A has also been reported in human melanoma cell lines, although this activity is counteracted by the secreted antagonist Follistatin (FST-1) (Stove et al., 2004). Immunohistochemical analysis detected Activin-A staining in superficially spreading melanoma, whereas benign nevi and metastatic lesions showed elevated expression of Follistatin (Heinz et al., 2015). Interestingly, however, gain of transgenic Activin-A expression in the A375 human melanoma xenograft model did not alter tumor growth or metastasis, even though it reduced tumor lymphangiogenesis, a risk factor of poor prognosis (Heinz et al., 2015). Also in xenograft models of other cancers, overexpression of a dominant negative mutant receptor revealed no adverse functions for Activin-A or related ligands, with the exception of tumor-induced systemic weight loss (cachexia) through remote Smad2,3 activation in skeletal muscles (Li et al., 2007, Zhou et al., 2010.
Here, we asked whether a potential tumorigenic role for Activin signaling in melanoma may be curtailed in xenografts by the absence of a functional immune system. To address this, we monitored the expression of INHβA in public datasets and in a panel of melanoma tumors and cell lines, and we performed loss-and gain-of-function studies, respectively, in human xenografts and in the moderately metastatic syngeneic B16F1 mouse melanoma model using dominant negative or ligand-independent mutant Activin/Nodal receptors or Activin-A lentiviral transgenes. Our comparison of melanoma grafts in tumor hosts of different genetic backgrounds reveals a novel protumorigenic and prometastatic function specifically for paracrine Activin-A signaling acting on adaptive immunity, and that this function can be uncoupled from immuneindependent anorexic and proangiogenic effects and from autocrine effects on melanoma cell differentiation. Our findings suggest that gain of Activin-A expression in human melanoma should be considered as a potential new target for improved immunotherapies.

Human melanoma and other skin cancers frequently express Activin-A rather than NODAL
To assess potential roles of INHβA in melanoma, we queried a gene expression profiling dataset of 42 primary cutaneous tumors (Riker et al., 2008). We found that INHβA expression was significantly elevated both in primary cutaneous cancers and in metastatic melanoma, whereas the antagonist Follistatin tended to decrease in melanoma in situ (n=2, not significant) compared to normal skin. In contrast to INHβA, other Activin receptor ligands encoded by NODAL or GDF3 were neither upregulated in this dataset nor in a TCGA collection of 474 skin cutaneous melanoma (Figs. 1A, S1). antagonist Follistatin in the TCGA dataset was associated with worse outcome (Fig. S3).
These results are consistent with a potential role for Activin-A signaling in melanoma progression.

Inhibition of endogenous ActRIIB ligands in human C8161 melanoma cells inhibits cachexia but not tumor progression in immunocompromised nude mice
Previous analysis of Activin-A functions in tumor xenograft models uncovered potent anorexic activity but no important roles in tumor progression (Heinz et al., 2015, Zhou et al., 2010. Instead, increased aggressiveness in human melanoma has been attributed to a secreted Nodal-like activity that was initially detected by injecting C8161 melanoma cells into zebrafish embryos (Topczewska et al., 2006). Since we observed no NODAL expression in human melanoma, we decided to reassess how C8161 cells stimulate Nodal/Activin receptors in zebrafish. To address this, we grafted C8161 cells into wildtype or mutant zebrafish embryos lacking maternal and zygotic expression of the Nodal co-receptor one-eyed pinhead (MZoep) (Gritsman et al., 1999). To validate that human NODAL signaling requires Oep in zebrafish, we injected 50 pg of mRNA encoding human Nodal at the one-cell stage and monitored induction of the Smad2,3 target gene gsc during early gastrulation. We found that both human Nodal and the zebrafish homolog Squint were active in wild-type but not in MZoep, whereas human Activin-A induced gsc independently of Oep (Fig. S4A). Moreover, both wild-type and MZoep embryos ectopically upregulated gsc as well as ntl around grafted C8161 melanoma cells (Fig. S4B). These results strongly argue for secreted Activin-A and against Nodal as the mediator of Smad2,3 signaling secreted by C8161 melanoma grafts in zebrafish.
To test whether melanoma may grow faster when forced to express Nodal, C8161 cells transduced with Nodal lentivirus (Fig. S2D) were grafted subcutaneously into FoxN1 nu/nu mice. C8161 xenograft tumor growth was not increased by Nodal compared to cells transduced with empty vector, indicating that it was not limited by lack of Nodal expression ( Fig. S4E, F). To assess the function of endogenous ligands in melanoma xenografts, we transduced human C8161 cells with lentivirus expressing an Fc fusion of Activin type IIB receptor extracellular domain (AIIB-Fc) or Fc alone. Lentivirally transduced AIIB-Fc inhibited endogenous Activin activity in conditioned medium of C8161 cells in vitro and was readily detected by immunoblotting in C8161 tumor xenografts (Fig. 1D, E). As expected, expression of AIIB-Fc in C8161 xenografts significantly protected FoxN1 nu/nu hosts against loss of body weight and muscle mass ( Fig. 1F). However, despite potent systemic inhibition of cachexia, soluble AIIB-Fc receptor neither diminished intradermal tumor growth nor experimental lung metastases after tail vein injection (Fig. 1G, H). Additionally, secondary tumor outgrowths following resection of the primary C8161 graft and spontaneous lung or lymph node metastases were not significantly inhibited ( Fig. S5A-C), nor did AIIB-Fc transduction inhibit experimental lung metastases of Me343 cells which also express INHβA (Figs. 1I, S2G). These results suggest that C8161 xenografts grow and metastasize independently of Activin signaling and of associated systemic cachexia, possibly because of the lack of functional adaptive immunity.

Activin-A gain of function promotes phenotype switching of mouse melanoma cells and tumor progression
To evaluate potential tumorigenic or prometastatic effects of Activin-A in immunocompetent hosts, we transduced INHβA in the moderately metastatic syngeneic B16F1 mouse melanoma model (Fidler, 1973) using lentiviruses for GFP alone (CTRL) or GFP together with Activin-A (INHβA). Western blot analysis of B16F1-conditioned medium and cell lysates confirmed Activin-A secretion which increased the accumulation of pSmad2 compared to CTRL cells that do not secrete Activin-A ( Fig. 2A). Transfection of the SMAD3 luciferase reporter CAGA-Luc showed elevated autocrine Activin-A signaling in INHβA-compared to CTRL-transduced B16F1 cells, and this increase was blocked upon treatment with Follistatin (Fig. 2B, left panel).
Conditioned medium from INHβA-transduced B16F1 cells also stimulated the expression of transfected CAGA-Luc in HEK293T reporter cells (Fig. 2B, right panel).
Furthermore, gain of Activin-A expression in B16F1 cells stimulated their migration in a scratch wound assay and markedly reduced their melanin content without altering their cell proliferation or viability ( Fig. 2C-E), consistent with a potential autocrine function in promoting a phenotypic switch.
To determine if oncogenic effects of Activin-A gain-of-function may promote tumor growth in vivo, B16F1 grafts were intradermally inoculated on the right flank of 8-10 week old female C57BL/6 mice. Mice injected with INHβA-transduced cells grew significantly larger tumors, combined with loss of body weight and cardiac muscle wasting ( Fig. 2F-H). Compared to vector control, INHβA also increased the number of experimental lung metastases formed by tumor cells that were injected into the tail vein ( Fig. 2I). Taken together, these results show that Activin-A signaling can promote melanoma growth and metastasis in immunocompetent hosts.

Sustained cell autonomous ALK4 signaling inhibits B16F1 tumorigenesis rather than stimulating it
To validate whether the tumorigenic function of Activin-A involves cell nonautonomous paracrine signaling, we transduced B16F1 cells with lentivirus expressing a constitutively active ligand-independent truncated form of ALK4 (caALK4). Induction of caALK4 by doxycycline led to increased Smad2,3 phosphorylation and expression of the SMAD3-dependent luciferase reporter CAGA-Luc ( Fig. 3A-C) and reduced pigment secretion (Fig. S6A). However, caALK4 expression decreased below detection within four subsequent cell passages despite continuous presence of doxycycline (Fig. 3A). induction in vitro was reduced (Fig. 3D), indicating impaired cell viability. A similar reduction in the number of viable cells was induced by ligand-independent caALK4 in human C8161 melanoma cells. To investigate the impact on tumor growth, B16F1 cells transduced with caALK4 were grafted intradermally into syngeneic C57BL/6 hosts, followed by treatment with doxycycline or empty vehicle until the endpoint of the experiment. Tumors induced no overt loss of body weight (Fig. 3E), and they grew significantly less in doxycycline-fed animals than in vehicle-treated controls (Fig. 3F), indicating that sustained autocrine Activin receptor signaling does not stimulate but rather suppresses tumor growth in vivo. Immunostaining at the endpoint of the experiment revealed increased apoptosis marked by cleaved Caspase-3 (Figs. 3G, S7E) while Ki67-stained proliferating cells appeared to decrease, albeit not significantly ( Fig. 3H-I and Supp. Fig. S7F). We conclude that while paracrine functions of Activin-A promote cachexia and tumor growth, sustained autocrine receptor signaling in vivo reduces B16F1 cell survival and tumorigenicity.

Activin-A-dependent B16F1 melanoma growth is mediated by impaired tumor immune surveillance
To test whether paracrine Activin-A signaling promotes tumor progression by inhibiting anti-tumor immunity, B16F1 cells were inoculated subcutaneously or intravenously into FoxN1 nu/nu that lack functional T cells, or into Rag1 -/mice devoid of V-D-J recombination in antigen receptors. In sharp contrast to immune-competent wild-type hosts, loss of either FoxN1 or Rag1 enabled CTRL tumors to grow as fast as those with sustained INHβA expression, even though Activin-A provoked cachexia irrespective of the host genotype ( Fig. 4A-F). These results suggest that Activin-A accelerates B16F1 melanoma growth by blunting T cell-mediated anti-tumor immunity. In good agreement, the same lentiviral INHβA transgene also failed to stimulate tumor growth or experimental lung metastases in xenografts of human melanoma cell lines in immunodeficient FoxN1 nu/nu mice ( Fig. S5D-G).

Angiogenesis is enhanced by Activin-A but is not sufficient to promote B16F1 melanoma growth
Depending on the context, Activin-A signaling may also promote or inhibit angiogenesis which can be rate-limiting for tumor oxygenation and nutrient supply (Lewis et al., 2016). To assess whether increased tumor growth correlates with increased tumor vascularization, we labelled blood vessels in thick cryosections of syngeneic B16F1 grafts using CD31 antibodies. Quantification in z-stack reconstructions of entire sections of 15 CTRL and 14 INHβA tumors showed that Activin-A significantly increased the vascular density (Fig. 5A). Conversely, pimonidazole staining of hypoxic areas in the same sections was 4-fold reduced, indicating that Activin-induced blood vessels were functional (Fig. S8A, B). Since Activin-A has been shown to inhibit endothelial cell growth and tubule formation in vitro (Kaneda et al., 2011), we asked whether its pro-angiogenic effect in B16F1 melanoma could be mediated by macrophages. However, FACS sorting of dissociated B16F1 melanoma at the experimental endpoint revealed no changes in total myeloid populations, and M2-like macrophages marked by CD206 staining were decreased in INHβA compared to CTRL tumors (Fig. S8C). Also in human melanoma with the highest levels of Activin-A, markers of lymphocytic or myeloid infiltrates appeared to be reduced rather than increased (Fig. S9), although recruitment of such infiltrates or their functions may vary depending on Activin-A signaling strength or duration. Interestingly, angiogenesis was similarly increased in Rag1 -/hosts were tumor growth remained unchanged upon INHβA overexpression, suggesting that increased vascularization alone cannot account for the tumorigenic effects of Activin-A (Fig. 5B).

DISCUSSION
Previous studies reported both tumor-suppressive and oncogenic effects of Activin-A, but a role in anti-tumor immunity and the relative contributions of autocrine versus paracrine signaling in an in vivo model of melanoma remained to be evaluated. Here, we show that the net outcome of a gain in paracrine Activin-A signaling in mouse and human melanoma grafts is determined by whether or not the tumor host has functional adaptive immunity. Tumorigenic paracrine effects on adaptive immunity trumped proapoptotic autocrine signals within cancer cells to overall facilitate primary and metastatic growth. Ectopic Activin-A signaling also stimulated tumor vascularization and, concurring with previous reports, systemic cachexia, and these effects were preserved in Rag1 -/mice albeit without accelerating tumor growth. Thus, a potential boost in nutrient supply by recycled tissue breakdown products was either insufficient to fuel tumorigenesis or neutralized by growth-inhibitory autocrine Activin receptor signaling within tumor cells. To our knowledge, these results furnish the first direct evidence that adaptive immunity is required for a tumorigenic Activin function, and that autocrine and paracrine signaling mediate opposite effects on melanoma growth in vivo. Based on these findings, future strategies to boost the efficacy of immunotherapies should consider targeting Activin-A.

Activin-A only stimulates melanoma growth in mice that have functional T cells
Our main finding is that paracrine Activin-A signaling was tumorigenic specifically in immunocompetent hosts, despite a proapoptotic function of autocrine Activin/Nodal receptor signaling within melanoma cells. Thus, a lentiviral INHβA transgene encoding secreted Activin-A in syngeneic B16F1 mouse melanoma grafts increased intradermal tumor growth and the frequency of experimental lung metastases specifically in wildtype C57BL/6 mice. By contrast, in syngeneic Rag1 -/hosts that lack T-and Blymphocytes, or in athymic  (Antsiferova et al., 2011). Depletion of CD4-positive T cells, including Tregs did not suppress the tumorigenicity of transgenic Activin-A in this skin carcinoma model (Antsiferova et al., 2016). However, it will be interesting to compare in future studies the potential of Tregs or tumor-infiltrating cytotoxic T-lymphocytes or other T cell subsets to mediate effects of Activin-A on anti-tumor immunity in melanoma. A role in suppressing anti-tumor immunity may explain why INHβA expression is more frequently upregulated in human melanoma and other solid human tumors than expected for a neutral bystander (Hoda et al., 2016, Wu et al., 2015.

Inhibition of melanoma growth by autocrine signaling may curtail oncogenic effects of secreted Activin-A on the tumor microenvironment
Even though Activin receptor signaling inhibits cell proliferation and induces apoptosis in normal melanocytes (Stove et al., 2004), treatment with recombinant Activin-A did not inhibit the proliferation or viability of cultured B16F1 cells in vitro. To more directly assess a potential role for autocrine Activin/Nodal signaling in the B16F1 model, we introduced a doxycycline-inducible ligand-independent mutant ALK4 transgene. We found that autocrine ALK4 signaling inhibited B16F1 cell proliferation and tumor growth rather than stimulating it. This confirms that oncogenic Activin-A activity in immunocompetent syngeneic hosts was mediated by paracrine signaling. Since melanoma cell survival was impaired by caALK4 but not by secreted Activin-A, autocrine signaling activity of the ligand is likely attenuated. In the hepatocyte lineage, autocrine anti-proliferative Activin-A signaling is frequently attenuated by secreted antagonists in hepatocarcinoma (reviewed in Deli et al., 2008). Also in human melanoma, dynamic changes in the levels of FST expression during progression of melanoma in situ to metastatic growth may modulate Activin-A responses (Heinz et al., 2015, Stove et al., 2004. However, whether FST influences the balance between autocrine and paracrine Activin signaling remains to be determined.

Interplay of tumor growth and cachexia
Commonly associated with advanced cancer in human patients, cachexia reduces life quality and drug responses while increasing morbidity and mortality (Fearon et al., 2013). In pancreatic MIA PaCa-2 xenografts, a comparison of tumor growth with the time course of Activin-induced cachexia suggests that associated metabolic changes or general weakening curtails a growth-promoting effect of autocrine Activin signaling (Togashi et al., 2015). If cachexia similarly curbs the growth of B16F1 melanoma growth, tumor growth should slow down concurring with the onset of Activin-induced cachexia at least in immunodeficient hosts. Such a trend in Rag1 -/hosts was not statistically significant and not seen in nu/nu hosts. The observed tumor growth curves also do not support a model that cachexia is rate-limiting for the supply of essential amino acids and other metabolites to cancer cells in a process of 'auto-cannibalism' (Theologides, 1979).

Activin-induced tumor angiogenesis and its uncoupling from tumor growth
We found that paracrine Activin-A signaling in the syngeneic B16F1 melanoma model also stimulated tumor vascularization. However, a similar increase of blood vessels in Rag1 -/hosts was not sufficient to facilitate tumor growth. Although we could not stain enough tumors in Rag1 -/mice with pimonidazole to formally exclude a stimulatory effect of adaptive immunity on vessel functionality and hypoxia, it is interesting to note that tumor growth was also uncoupled from angiogenesis in immunodeficient SCID mice bearing mammary carcinoma xenografts, where gain of Activin-A signaling diminished tumor angiogenesis without affecting vascular perfusion (Krneta et al., 2006). Interestingly, however, Activin signaling within endothelial cells in this breast cancer model and other tumors was cytostatic and reduced sprouting and blood vessel density (Breit et al., 2000, Kaneda et al., 2011, Krneta et al., 2006, indicating that proangiogenic activity is likely indirect. T cell-derived cytokines are unlikely involved since Activin-A stimulated the vascularization of B16F1 tumors even in Rag1 -/syngeneic hosts. Possibly, class I inflammatory macrophages that stimulate vascularization in Activin-induced skin squamous cell carcinoma mediate proangiogenic activity (Antsiferova et al., 2016). Although, the total number of infiltrating macrophages was unchanged by Activin-A in B16F1 melanoma under the conditions examined, and since increased angiogenesis was not sufficient to promote tumor growth, we did not further investigate whether Activin-A directly polarized a proangiogenic macrophage subtype in this model.
Overall, our findings suggest that paracrine Activin-A should be considered as a novel target for personalized therapies not only to reduce cachexia and melanoma vascularization, but also to enhance the efficacy of immunotherapies.

Melanoma grafts
1x10 6 B16F1 cells were injected intradermally into the right flank of 8-12 week old female wild-type (Harlan) or Rag1 -/-C57BL/6 (EPFL animal core facility) syngeneic hosts, or of Hsd-athymic nu/nu mice (Harlan). Animal body weights and tumor sizes were measured every two days. Tumor volumes were calculated using the formula length x width x depth. Experimental lung metastases were obtained by injecting 3x10 5 B16F1 cells into the tail vein. Pigmented metastases visible at the surface of each lung lobe were counted 3 weeks after injection. Where indicated, animals were fed with chow containing 0.625 g/kg doxycycline (Provimi Kliba AG, Switzerland). All procedures were according with Swiss legislation and approved by the cantonal veterinary adminstration. Generation of cell lines and in vitro assays are further documented in Supplemental Methods.

Gene expression analysis
Total RNA from melanoma cell lines and tumors was isolated using Trizol reagent (Sigma or ThermoFisher) and guanidinium/CsCl gradient, respectively, or by using RNeasyMini kit (Qiagen). 1 µg of total RNA was used for cDNA synthesis using the Superscript III Reverse Transcription Kit (Invitrogen). Quantitative polymerase chain reaction (qPCR) assays were performed using SYBR® green chemistry according to manufacturers' instructions (Applied Bioscience), or commercial Taqman probe for INHβA (Eurogentec). PCR primer sequences are listed in Table S1.

Cell cycle and Ki67 analysis
After treatment with 250 µg/ml doxycycline for 24, 48 or 72 hours in 6-well plates, cells were incubated with 10 µM EdU for 30 min and then trypsinized, washed with PBS, and fixed for 20 min with 4% PFA, permeabilized with 0.5% Triton X-100 in PBS and stained for 30 minutes with Alexa-647-coupled azide in presence of copper sulfate and sodium ascorbate. Cells were then washed and stained with DAPI and analyzed using a

Statistical analysis
All statistical analyses were performed using GraphPad Prism v6 software. Data were analyzed using Mann Whitney (for non-parametric data), T-tests, 1-way or 2-way ANOVA with Bonferroni correction for parametric data. A p-value <0.05 was considered significant. animal care facility. This work was supported by grants to D.B.C. from the Swiss Cancer  I) As in H), but using human Me343 melanoma cells (n=10 per group, p=0.14)        tumor growth compared to control cells when grafted into immunodeficient nu/nu mice (Postovit et al., 2008, Topczewska et al., 2006. However, a 35 kDa protein detected in extracts of C8161 melanoma cells (Hardy et al., 2010, Postovit et al., 2008, Topczewska et al., 2006 did not correspond in size to the known molecular weights of NODAL precursor (39-41 kDa) or processed form (12 kDa) Robertson, 1999, Le Good et al., 2005). Among potential splice variants (Strizzi et al., 2012), only two transcripts share exons 2 and 3 encoding a functional NODAL precursor that can react with antibodies against the mature region (Fig. S2A). Therefore, to monitor NODAL mRNA expression, we designed intron-spanning primers in exons 2 and 3 which detect correctly spliced mRNAs in fetal brain or glioblastoma control samples. Unexpectedly, however, we did not detect spliced NODAL transcripts in C8161 cells or in any of the melanoma cell lines or 24 human melanoma biopsies analyzed, including C8161 cells, even after as many as 35 PCR cycles (Figs. 1B, S2B). Others have argued that correctly spliced NODAL mRNA can only be detected with special technical expertise using primer sequences that were not disclosed (Strizzi et al., 2012). However, a technical problem with our primers is unlikely since they amplified spliced NODAL mRNA fragment of the correct size (288 bp) both in fetal brain and glioblastoma total RNA ( Fig. S2B).

A-B) Western blot of A) secreted Activin-A and B) CAGA-Luc expression in
To analyze NODAL protein, we raised a custom antibody against the fully conserved peptide KQYNAYRCEGECPNPV of human NODAL (Fig. S2C). In Western blots of transfected HEK293T and C8161 cells transduced with Nodal lentivirus (positive controls), the custom antiserum reacted with these recombinant control proteins as expected. However, no specific band was detected either in fresh lysates of nontransfected C8161 cells or in the less aggressive C81-61 cell line or their conditioned media (Fig. S2D, and data not shown). Species differences of human NODAL cannot account for the absence of specific bands since mature mouse Nodal is 98% identical and the antigenic peptide is 100% conserved (Fig. S2C). Also in the prodomain, NODAL is only 6 amino acids shorter in human than in mouse, and a single N-glycosylation motif (NWT) that is functional in mouse Nodal (Blanchet et al., 2008) is conserved.
To address the discrepancy between these results and published data, we compared our custom antibody with the commercial antibodies from Santa Cruz or R&D that were previously used to detect NODAL in melanoma (Hardy et al., 2010, Topczewska et al., 2006. While both of them recognized recombinant Nodal in HEK293T cells, they were less sensitive than our custom antiserum both on stripped (Fig. S2D) and on non-stripped blots (not shown). Furthermore, superposition of blots revealed that each commercial antibody also cross-reacted with distinct sets of additional bands around 35 kDa and above 49 kDa, especially if cells were extracted with detergent (Postovit et al., 2008) instead of Laemmli buffer (Fig. S2E). Importantly, however, we also observed these bands in MyoEP cell extracts (negative control, courtesy of Dr. Postovit), and proteins reacting with the Santa Cruz antibody did not overlap with those detected by R&D antibody (Fig. S2E). Therefore, and since no endogenous NODAL was detected with the more sensitive custom Nodal antibody, it seems that both commercial antibodies each cross-react with distinct non-specific proteins.
To distinguish whether C8161 cells secrete NODAL or a related activity, we tested the potential of conditioned medium to induce the SMAD3 luciferase reporter CAGA-Luc in HEK293T cell lines that express the Nodal co-receptor Cripto. As a negative control, we also used analogous reporter cells stably transduced with the related protein Cryptic. As shown previously (Fuerer et al., 2014), only Cripto but not Cryptic mediate induction of luciferase expression by Nodal, whereas treatment with recombinant Activin-A potently induced the reporter in both cell lines (Fig. S2F). Importantly, CAGA-Luc was also induced by C8161-conditioned medium independently of Cripto, and this activity was sensitive to inhibition by the Activin-specific antagonist Follistatin which does not inhibit NODAL (S2F). Concurring with the above RT-PCR and Western blot data, these results indicate that C8161 cells do not secrete detectable amounts of Nodal activity under the conditions examined. To directly validate whether C8161 secrete Activin-A, we also analyzed cell lysates and conditioned medium by Western blot. Using an antibody that reacts only with secreted Activin-A dimers, we detected mature Activin-A in its processed form specifically in conditioned medium of C8161 and Me300 cells, but not in the medium of C81-61 or Me290 melanoma cells or in cell lysates (Fig. S2G).
Taken together with our zebrafish experiments (Fig. S4), these results show that C8161 cells under the conditions examined secrete biologically active Activin-A but not detectable amounts of NODAL. and with Cripto or Cryptic have been described (Fuerer et al., 2014). All cell lines were tested before in vivo use and found to be free of Mycoplasma (SouthernBiotech, 13100-01).
For cell transplantations, 50-100 melanoma cells (≈500 pl of a cell suspension with 10 8 cells/ml in Ca 2+ -free Hank's medium) were grafted into pronase-dechorionated wild-type (TE) or MZoep embryos at sphere stage as described (Topczewska et al., 2006). After incubation at 31°C for 3 hours until shield stage, embryos were fixed and stained for ntl and gsc by whole mount in situ hybridization (Muller et al., 2012). Embryos were imaged using an Axio Zoom.V16 microscope (ZEISS).

Reporter assays
HepG2 CAGA-Luc reporter cells stably expressing the Nodal co-receptor Cripto as well as Renilla luciferase for signal normalisation (Fuerer et al., 2014)  were extracted for 20 min in 100 µl potassium phosphate buffer containing 0.5% Triton X-100, and luciferase activities

Scratch wound assay
Cells were plated in 6-well plates at confluent density of 4x10 6 per well. After letting cells adhere for six hours, a scratch wound was made in the cell monolayer using a P200 pipette tip, and the place of the scratch was marked. The cells were washed with PBS and the medium was changed for the indicated treatments. Scratches were imaged at 0 hours and after 20 hours.